Win10 GPU Target from Linux Host

This ought to be straightforward, but I am apparently missing the secret recipe. I have a Windows 10 machine with two NVidia GPUs installed plus a full CUDA 8.0 development environment, and I’m running the NSight Monitor as administrator. I have a CentOS 7 Linux machine, also with a full CUDA 8.0 development environment, but no resident GPUs. I would like to use NSight Eclipse on the Linux machine, and debug GPU applications using the GPUs on the Windows machine. Nothing I have tried so far has worked; ultimately, I cannot successfully connect via SSH from NSight to Windows, even though I can SSH from the Linux command line just fine. I’ve tried configuring a target as a Windows target, but am not managing to get the overall process working. Documentation is light-to-non-existent for this approach, and I haven’t had much luck with Google or the forums. I would appreciate any insight (no pun intended) that anyone could offer. And please, do not tell me to instead use NSight Visual Studio on the Windows machine; yes, that obviously works, but I have other objectives requiring the Linux platform as the host.

Hi, Mr_Software

I’m afraid we do not support this scenario.
For now, nsight EE only support target as Linux or Android, but not Windows.

Actually, target requires ‘cuda-gdbserver’ to work at least for remote connection, but we do not have this released for windows.

Sorry for the inconvenience.

Veraj,

Thanks for the prompt reply, even if it is bad news. I would also assume that Windows System Linux is likely not an alternative, as it may not have sufficient access to the GPU, correct?

Mr_Software

I think it is not GPU access issue caused this. Just the implementation do not support.

Like we still can use nvvp(Visual Profiler) in windows and remote profile on Linux GPU.

Mr_Software

I think it is not GPU access issue caused this. Just the implementation do not support.

Like we still can use nvvp(Visual Profiler) in windows and remote profile on Linux GPU.